IgnisPrompt
Local-first routing for routine AI tasks + smart escalation to LLMs + personal impact dashboard that makes sustainability visible and rewarding.
Your Personal Impact Dashboard
The most exciting part of IgnisPrompt isn't just that it routes smarter — it's that you can see and feel the impact. Like a health app tracks your steps toward a fitness goal, IgnisPrompt shows your contribution to reducing AI waste.
Dashboard Mockup Goes Here
(Replace with Figma export showing: route trends, cost savings graph, emissions tracker, goal progress rings)
All estimates use transparent assumptions (IEA grid intensity, data center PUE) that you can override with your own data.
Handles routine AI work locally, escalates intelligently
Most AI tasks are lightweight: summarizing a short message, extracting dates, drafting a quick reply, reformatting text. IgnisPrompt routes these to small local models and saves cloud LLMs for work that truly needs them.
Detect intent & complexity
Classify the request: Is it summarization? Extraction? Generation? Simple or complex? Fast intent detection with configurable thresholds.
Route to the right model
Routine work → local/edge small models. Complex reasoning → cloud LLMs. Sensitive data → on-device models with no network call. Policy-based routing for compliance and cost control.
Collect context before escalating
Before calling a cloud LLM, IgnisPrompt gathers all necessary data: user context, relevant documents, structured inputs. One well-prepared request beats three back-and-forth calls.
Measure everything
Every decision is logged: which route, which model, how long, estimated cost, estimated energy use. Export reports, track trends, optimize thresholds based on real data.
Most daily AI work is lightweight
These tasks don't need GPT-4 or Claude Opus. A well-tuned small model running locally can handle them instantly, with no network latency and no cloud inference cost.
Common Local-First Tasks
- Summarization: Condense short emails, messages, meeting notes
- Extraction: Pull dates, names, action items, key facts from text
- Simple drafting: "Reply saying yes", "Write a polite decline", "Format this as bullet points"
- Classification: Ticket triage, sentiment, intent detection
- Reformatting: Convert prose to list, markdown to plain text, etc.
- Translation (common pairs): English ↔ Spanish, French, etc. for short text
- Quick Q&A: Factual lookups from a known document set
When does IgnisPrompt escalate to cloud LLMs?
When the task needs deep reasoning, long context, specialized knowledge, or creative generation. When confidence is low. When the user explicitly requests a cloud model. When policy requires external validation. Escalation is smart, not dogmatic.
What you can measure with IgnisPrompt
We show our assumptions, let you override them, and make results reproducible. No greenwashing, no vague claims.
Direct Signals (measured)
- Route decision: Local, edge, or cloud for every request
- Latency: p50, p95, p99 response times
- Tokens: Input/output size where available
- Model class: Which model handled the request
- Success rate: Quality checks, user feedback
Estimates (transparent assumptions)
- Cost savings: Based on cloud model pricing vs local compute
- CO₂ reduction: Grid intensity (IEA baselines), data center PUE, token→kWh conversion
- Water footprint: Indirect (power generation) + direct (cooling)
- All assumptions documented: Export methodology + overrides panel
Ready to see IgnisPrompt in action?
We're looking for design partners to pilot IgnisPrompt on real workloads. Be among the first to shape the product.